Hugging Face's logo

Datasets:
THULab
/
cgtsf

Tasks:
Time Series Forecasting
Modalities:
Time-series
Size:
10K<n<100K
ArXiv:
Tags:
tsfile
timeseries
modality:timeseries
format:tsfile
multimodal-time-series
context-guided-time-series-forecasting
License:
Dataset card Files
xet
 Community
cgtsf / README.md
wangdx25's picture
wangdx25
Add TsFile (converted from ChengsenWang/CGTSF)
012b018 verified
preview code
|
Raw History Blame Contribute Delete
3.77 kB
metadata 
license: apache-2.0
task_categories:
  - time-series-forecasting
size_categories:
  - 10K<n<100K
tags:
  - tsfile
  - timeseries
  - modality:timeseries
  - format:tsfile
  - multimodal-time-series
  - context-guided-time-series-forecasting
  - energy
  - traffic
modality: timeseries
pretty_name: CGTSF Context-Guided Time Series Forecasting (TsFile format)
configs:
  - config_name: leu
    data_files:
      - split: train
        path: cgtsf_leu.tsfile
  - config_name: mspg
    data_files:
      - split: train
        path: cgtsf_mspg.tsfile
  - config_name: ptf
    data_files:
      - split: train
        path: cgtsf_ptf.tsfile

CGTSF: Context-Guided Time Series Forecasting (TsFile format)

Apache TsFile version of ChengsenWang/CGTSF.

Modalities: Time-series.

Overview

CGTSF addresses context-guided time-series forecasting by aligning numerical history and prediction windows with textual background, weather, and date information. The source provides three logically distinct multimodal domains: London electricity usage (LEU), Melbourne solar power generation (MSPG), and Paris traffic flow (PTF). The source card reports the following frequencies and series counts:

Config Rows Series Frequency Array widths (Hist / Pred)
LEU 11,616 16 30 minutes 240 / 48
MSPG 10,557 27 15 minutes 480 / 96
PTF 11,520 32 1 hour 120 / 24

The three source files remain separate because their serialized array widths are incompatible. The source card describes PTF as a 2012 traffic dataset, while the observed source dates at this pinned revision are in 2022; the conversion follows the source date values without changing them.

Schema (TsFile structure)

Each config contains its own table with the same structural roles:

  • Time (INT64, milliseconds) - the source target-day Date at UTC midnight.
  • idx (TAG, STRING) - Idx, identifying the source series/device.
  • hist_000 ... hist_N (FIELD, DOUBLE) - the serialized Hist numeric window expanded to scalar fields; N depends on the config.
  • pred_000 ... pred_N (FIELD, DOUBLE) - the serialized Pred window expanded to scalar fields; N depends on the config.
  • text (FIELD, STRING) - the aligned textual context.

Hist and Pred are parsed with a safe literal/JSON parser and expanded without changing their numeric values. The source Date is represented losslessly by Time; no source row or semantic column is dropped.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("cgtsf_leu.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license